Distributed filtering and local statistical approach to fault diagnosis for securing the power grid

Distributed Kalman filtering and the local statistical approach to fault diagnosis are used to develop a systematic method for the detection of attacks against sensors of the power grid. To treat the case of grid's frequency may deviation from its nominal value as well as the case of imprecise grid frequency measurement (i) the sensors dynamics is described using multiple local Kalman filters associated with different measurements of the grid's frequency, and (ii) fusion of the local filters' estimates is performed, in the sense of distributed Kalman filtering, so as to obtain a reliable aggregate estimate of the sensors' state. Next, to emulate the functioning of the grid's sensors in the fault-free mode, the distributed Kalman filter is used as a virtual sensor. By comparing the output of the distributed Kalman filter against the output of the real sensors, the resulting differences generate the residuals' vector. The residuals sequence undergoes statistical signal processing, so as to determine if specific sensors have undergone an intruder;s attack. The generalized likelihood ratio of the residuals sequence is used, as a statistical change detection criterion. This provides a statistical test that is based on the χ distribution, and allows to detect deviation of the sensors' functioning from normal mode. Using the properties of the χ distribution, an optimal threshold is defined for deciding on whether a sensor has been providing distorted measurements or not. Besides, with the application of this statistical criterion to clusters of sensors within the entire sensors' set it is possible to isolate those particular sensors which have been exposed to failure or an intruder's attack. The method achieves detection of sensors' malfunctioning that differs less than 1% from the nominal sensor's output. The application of the proposed method contributes to the enforcement of the security levels of the electric power grid.